l 1 intro machine learning
TRANSCRIPT
-
8/12/2019 l 1 Intro Machine Learning
1/45
Unit Coverage
Machine Learning Basics
Machine Learning Applications
Supervised and Unsupervised learnin
Gradient Descent For learningCopyright @ gdeepak.com6/4/2014 6:53 PM
-
8/12/2019 l 1 Intro Machine Learning
2/45
Machine Learning
Machine learning, a branch of artificial intelligenconcerns the construction and study of systems thatlearn from data. -Wikipedia
It is a type of AI that provides computers with the abilitylearn without being explicitly programmed. Machlearning focuses on the development of computer prograthat can teach themselves to grow and change wh
exposed to new data.
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
3/45
Machine Learning-Abstract Definitio
A Machine (Computer Program) Learns with experience E forSome Task T and Performance Measure P, if P keeps onincreasing with increase in E.
Example: Someone Writes a Program to classify and filter you
emails as spam or not based on your marking of individual maas spam or not. What is T, E and P in this example.
Your marking of an email as Spam or not
Percentage of emails being true positive for spam
Recording of your labelling or classification of email as spam
or not
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
4/45
Characteristic of Machine Learning
- Way learning happens is very critical- Applies to tasks that can not be defined well, except
examples
- To find relationships and correlations that can be hidd
in the data- To learn in proportion to the experience e.g. becoming
better player after playing many games
- Results may vary vastly if we apply different learn
paths or different algorithms of ML
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
5/45
Two Basic Types
Supervised LearningIn general it means that we are going to supervise
learning mechanism or we are going to supply someguidelines/ parameters/ labelling regarding the data
In supervised learning, training patterns giving inputs athe corresponding correct outputs are available.
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
6/45
Unsupervised Learning
Learning happens automatically and the structuhidden in the data are recognised by the system
System must find interesting and/or significant pattein the data without any feedback as to what is right
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
7/45
Grading Example
6/4/2014 6:53 PM Copyright @ gdeepak.com
A
B
C
D
Grades
Markss
-
8/12/2019 l 1 Intro Machine Learning
8/45
Handwriting Recognition
6/4/2014 6:53 PM Copyright @ gdeepak.com
S
S
S written by Different People
-
8/12/2019 l 1 Intro Machine Learning
9/45
Supervised Learning
6/4/2014 6:53 PM Copyright @ gdeepak.com
For each Data Point we will let the machine know whetheit is a star or smiley
-
8/12/2019 l 1 Intro Machine Learning
10/45
Classification or regression
When we have discrete outputs then the problem iclassification problem
When we have continuous outputs then it is a regressproblem
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
11/45
Traffic Time Prediction
You Give 10 Actual Timings to reach from AmbalaDelhi if you start at different times of the day starting9 A.M.
Now you want to Predict the timings at some other tim
You may use different Curve Fittings
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
12/45
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
13/45
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
14/45
Speech Recognition
Database of Requests User speaks something, you need to Identify the reque
You need to capture the individual recordings frmeeting of four people
You need t0 separate the conversation from backgrnoise or music
You need to understand the speech in a particular lanand do the text labelling
You need to convert the speech in some other languag
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
15/45
Medical Diagnosis
Given Symptom and Disease database A new patient with some symptom comes, you need to
identify the disease
To diagnose the disease from the test results and by
analysing the images from the medical equipment T0 recognise disability by looking at the photograph o
the person
Different machine learning tests for various disabilitie
e.g. hearing test
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
16/45
Unsupervised Learning
6/4/2014 6:53 PM Copyright @ gdeepak.com
Each Data Point is given but not labelled; machine is supposedfind some structure in the data; in this example called clusters
-
8/12/2019 l 1 Intro Machine Learning
17/45
News.google.com
All similar stories are clustered at one place
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
18/45
Participant Segmentation
If I give all the Registration Form data of IWMLDA to Sounsupervised learning based program and it comes out wsome grouping based on the distinguishing features giin the Registration Form.
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
19/45
Social Network Analysis
To find groups of certain kind on theNetwork based on their activity, Cohesiveness,
Type of Chatter, Type of likes etc
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
20/45
Sentiment Analysis
You want to buy a product and you want to knowsentiments of the public who has previously usedbought that product.
There can be different kinds of sentiments that hbeen expressed online; It may be related to spoPolitics, Tragedy, Agitation/ Revolution
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
21/45
Semi-Supervised Learning
Semi-supervised learning is a class of supervised learntasks and techniques that also make use of unlabeled dfor training - typically a small amount of labeled data witlarge amount of unlabeled data.
Actually it will depend upon the type and size of davailable.
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
22/45
Training Set-Old car price example
Mileage Car Price
2000 300000
20000 200000
18000 220000
100000 100000
50000 150000
80000 130000
10000 250000
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
23/45
Some Terminology
m= number of training records x= input features/ input values of the variables (can be
more than one)
y= output value ( Can be more than one)
(x,y) is a pair of one training record
(x (i), y(i)) is ith pair of training example
i is parameter of feature x
What is y4 and what is x2 on the previous slide
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
24/45
General Model
6/4/2014 6:53 PM Copyright @ gdeepak.com
Mileage (x)Hypothesis
(h)
Training Records
Car Price (y)
Learning
Algorithm
-
8/12/2019 l 1 Intro Machine Learning
25/45
Hypothesis Parameters
Only point to be kept in mind while selecting i is tha
should give the value of the hypothesis as close to y intraining record as possible
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
26/45
Cost Function using Squared Error Functi
6/4/2014 6:53 PM Copyright @ gdeepak.com
C Pl
-
8/12/2019 l 1 Intro Machine Learning
27/45
Contour Plots
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
28/45
Contour Figure
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
29/45
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
30/45
Another Example
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
31/45
Gradient Descent
To minimize the cost functionMin
(0, 1, 2.. n)J (0, 1, 2.. n)
Start with some initial values of
Keep applying gradient descent until we reach to theminimum possible value, which may be the optimal valuthe cost function.
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
32/45
Different Shape Bowls
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
33/45
For convergence
6/4/2014 6:53 PM Copyright @ gdeepak.com
Where alpha is the learning rate. Learning rate also pla
important role in the slow and fast convergence of GrDescent, but there is always a trade off. With small learningalgorithm may take many iteration and will be slow, whilelarge learning rate, the algorithm may be fast but it maconverge at all and we may skip or bypass the local or global m
All values of j f
to n shouldsimultaneouslyupdated
Concept of learning rate on bowl shape
-
8/12/2019 l 1 Intro Machine Learning
34/45
Concept of learning rate on bowl shapecurve
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
35/45
Gradient Descent with linear regressio
We repeat the following expression until the functionconverge for all values of .
6/4/2014 6:53 PM Copyright @ gdeepak.com
If each step of the the gradient descent uses all the training records then
algorithm comes under the category of batch gradient descent
-
8/12/2019 l 1 Intro Machine Learning
36/45
Dealing with multiple variablesMileage Car Price Engine Size
(No. of
Cylinders)
OriginalPrice
AccessorCost
2000 300000 4 500000 40000
20000 200000 6 600000 3000
18000 220000 4 450000 100000
100000 100000 4 400000 5000050000 150000 8 800000 110000
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
37/45
Feature scaling
Since the range of values of raw data varies widely, in somachine learning algorithms, objective functions willwork properly without normalization. For example,majority of classifiers calculate the distance between tpoints by the distance. If one of the features has a br
range of values, the distance will be governed by tparticular feature. Therefore, the range of all featushould be normalized so that each feature contribuapproximately proportionately to the final distance.
6/4/2014 6:53 PM Copyright @ gdeepak.com
-
8/12/2019 l 1 Intro Machine Learning
38/45
How to do feature scaling
Where is the mean or average value of the training valuof that feature and s is the range (max-min) of that featutraining value. We try to get every feature into
Range. However if the feature values are not too muchdistorted then we may not decide to go for feature scaling
6/4/2014 6:53 PM Copyright @ gdeepak.com
li l
-
8/12/2019 l 1 Intro Machine Learning
39/45
Feature Scaling Example
Average : 303000/5= 60600
Max-Min= 107000
6/4/2014 6:53 PM Copyright @ gdeepak.com
Mileage Car Price Engine Size(No. of
Cylinders)
AccessoryCost
AfterScaling
2000 300000 4 40000 -0.19
20000 200000 6 3000 -0.53
18000 220000 4 100000 +0.37
100000 100000 4 50000 -0.150000 150000 8 110000 +0.46
C bi i
-
8/12/2019 l 1 Intro Machine Learning
40/45
Combining Features
Few features may have same values but may have bgiven in different units. For ex. Height in cm and heighinches. Similarly few features have parallel values e.g lenof the string, number of characters in the string etc
6/4/2014 6:53 PM Copyright @ gdeepak.com
Other Imp points regarding Convergence
-
8/12/2019 l 1 Intro Machine Learning
41/45
Other Imp points regarding Convergence Gradient Descent
For small learning rate J() should decrease on eviteration of the algorithm.
Having learning rate too small or too large will haveown issues as discussed before.
The number of iterations may vary from two digitsmany digits.
If J() decreases by less than 0.001 then we can declconvergence, otherwise the delta change will besmall.
6/4/2014 6:53 PM Copyright @ gdeepak.com
Q i
-
8/12/2019 l 1 Intro Machine Learning
42/45
Question
Does the learning Rate remains same or it changes ovetime. If yes, why. If No, Why.
6/4/2014 6:53 PM Copyright @ gdeepak.com
Q ti
-
8/12/2019 l 1 Intro Machine Learning
43/45
Question
6/4/2014 6:53 PM Copyright @ gdeepak.com
Whether test sample (green circle)should be classified either to the firstclass of blue squares or to the secondclass of red triangles using k-NN
technique.Ifk = 3 (solid line circle)Ifk = 5(dashed line circle)
Q ti
-
8/12/2019 l 1 Intro Machine Learning
44/45
Question
What will be your criteria to decide whether to use featurscaling or not?
6/4/2014 6:53 PM Copyright @ gdeepak.com
Q ti S ti d C
-
8/12/2019 l 1 Intro Machine Learning
45/45
Questions, Suggestions and Commen
6/4/2014 6:53 PM Copyright @ gdeepak.com